Datasets:

OneScience-Group
/

Kolmogorov_flow_2d

license: apache-2.0
language:
  - en
  - zh
tags:
  - OneScience
  - Kolmogorov-flow
  - computational-fluid-dynamics
  - turbulence
  - vorticity
  - neural-operator
  - FactFormer
frameworks:
  - NumPy
pretty_name: Kolmogorov Flow 2D Re1000

Kolmogorov Flow 2D

Dataset Description

This dataset contains time series of single-channel vorticity fields obtained from numerical simulations of two-dimensional Kolmogorov Flow. It can be used for turbulence time-series forecasting, neural operator training, partial differential equation surrogate modeling, and long-horizon autoregressive forecasting.

The data describes scalar vorticity fields over a two-dimensional periodic domain. The dataset contains 120 trajectories, each with 320 frames, a spatial resolution of 256 x 256, a data type of float32, and a Reynolds number of Re = 1000. The dataset has been adapted for OneScience-Group/FactFormer.

Supported Tasks

Scenario	Description
Multi-step flow field forecasting	Predict flow fields for multiple subsequent time steps from historical vorticity fields.
Autoregressive time-series modeling	Feed model outputs back into the input window for long-horizon rollouts.
Neural operator research	Compare operator architectures such as Transformer, FNO, UNO, and KNO.
Turbulence surrogate modeling	Learn the spatiotemporal evolution mappings produced by numerical solvers.

Dataset Format and Structure

The main data file is in NumPy .npy format:

kf_2d_re1000_256_120seed.npy

The array shape is [120, 320, 256, 256], with the dimensions representing trajectory, time, x-grid, and y-grid, respectively. The original file is approximately 9.38 GiB. On-demand access via numpy.load(..., mmap_mode="r") is recommended to avoid copying the entire array into memory.

The data contains only vorticity fields; it does not include velocity, pressure, forcing fields, or physical time-step information.

How to Use the Dataset

Download the dataset:

hf download --dataset OneScience-Group/Kolmogorov_flow_2d --local-dir ./Kolmogorov_flow_2d

To use the dataset with FactFormer, set the data directory in FactFormer/conf/config.yaml to the download directory, and run:

cd FactFormer
python scripts/train.py
python scripts/inference.py

By default, FactFormer downsamples the spatial resolution to 128 x 128 and uses the first 10 frames to predict the subsequent 16 frames.

Official OneScience Information

Platform	OneScience Main Repository	Skills Repository
Gitee	https://gitee.com/onescience-ai/onescience	https://gitee.com/onescience-ai/oneskills
GitHub	https://github.com/onescience-ai/OneScience	https://github.com/onescience-ai/oneskills

Citation and License

Recommended model: OneScience/FactFormer.
FactFormer: Liu-Schiaffini et al., FactFormer: Factorized Transformer for Modeling Long-Range Dependencies in PDE Surrogate Modeling.
The documentation and supporting scripts in this repository are licensed under Apache-2.0. Before publicly distributing or redistributing the numerical data, verify its upstream source and licensing requirements.